fix(terminal): cross-platform compatibility for LocalEnvironment on Windows - #181
Closed
dogiladeveloper wants to merge 8 commits into
Closed
Conversation
Added Notion integration skill documentation including usage, setup, and procedures for various operations.
teddyjfpender
added a commit
to teddyjfpender/superforecasting-agent
that referenced
this pull request
Jun 28, 2026
The market-hidden arm ran (245 cases): the closed-book agent has NO intrinsic edge — Brier 0.279 vs market 0.164 (loses by 0.115), OVER-confident (SCE +0.20, worse than a base rate), simplex weight 0.000→0.031 but CI still includes 0 and beats_both=False. - Grounding study: new §3.4 (the head-to-head table + the over-confidence finding), abstract + §4 interpretation updated (the visible-arm skill was borrowed anchoring; the edge must be manufactured by the harness from fresh information, proven out-of-sample on future-resolving markets), §6.1 resolved (Gate-2 NousResearch#181 is the next step; de-extremize-when-uninformed noted). - Strategy §5: EMPIRICAL UPDATE recording that closed-book Lever A fails the falsification test — but it is the NO-SEARCH FLOOR, so it proves the precondition (parametric knowledge carries no edge) and makes Lever B (search, NousResearch#181) the ONLY lever, not merely the largest. Informed-A1 thesis still open.
teddyjfpender
added a commit
to teddyjfpender/superforecasting-agent
that referenced
this pull request
Jun 28, 2026
…fresh-search loop The market-hidden experiment proved the closed-book LLM has no intrinsic edge, so the ONLY path to beating the market is fresh information it hasn't priced. P1.1 built the run_quorum search seam; this wires a real one into the live quorum path. - NEW supervisor_search.py: build_supervisor_search_runner — a bounded (max 3 queries x 5 results, dedup, total cap 12) Callable that runs the agent's real web/news search for the judge's clarifying queries and reshapes rows into evidence dicts (available_at=now). Robust: a failing backend yields [] (never throws). Drops P2.4 leak-domain hosts so a settlement/ odds-widget URL can't be folded into the panel/judge context. - quorum_jobs.execute_job builds + passes the search_runner, GATED behind spec.supervisor_search (forecast quorum --supervisor-search) or the quorum.supervisor_search config flag; DEFAULT OFF => byte-identical (research_rounds 0). When on, the judge can trigger one fresh-search round. Review fix (MAJOR leakage seam): the search is now disabled in CODE for a HISTORICAL evidence_cutoff (_cutoff_is_live) — fresh present-day search cannot be pinned to a past cutoff, so a backtest/replay snapshot can never fold post-cutoff info into a past-pinned forecast (the no-leak invariant is enforced, not left to convention). Plus: clamp research rounds at 3, the leak-domain filter above, and corrected docstrings (the config flag governs the quorum path, not market-nightly). Full forecasting + quorum suite: 1556 passed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
abhibansal-sg
referenced
this pull request
in abhibansal-sg/hermes-agent-fork
Jul 4, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Problem
tools/environments/local.py uses three POSIX-only APIs that crash
on Windows with AttributeError:
CONTRIBUTING.md lists cross-platform compatibility as priority #2.
Changes
tools/environments/local.py
POSIX: os.getpgid + os.killpg(SIGTERM/SIGKILL), falls back to
proc.terminate()/proc.kill() on PermissionError
Windows: proc.kill() directly, no preexec_fn, no killpg
POSIX: [shell, "-lic", command] + preexec_fn=os.setsid (unchanged)
Windows: ["cmd.exe", "/C", command] — preexec_fn not passed
No behaviour change on Linux or macOS.
Testing
New file: tests/tools/test_local_environment.py
pytest tests/tools/test_local_environment.py -v
Checklist